tmpfs: optimize clearing when writing
authorHugh Dickins <hughd@google.com>
Tue, 29 May 2012 22:06:39 +0000 (15:06 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Tue, 29 May 2012 23:22:22 +0000 (16:22 -0700)
commitec9516fbc5fa814014991e1ae7f8860127122105
treee719b64b939e791499e115a63f4e7d1dd0448b16
parent2f6e38f3cd17a7858112f538c1700c747170db1f
tmpfs: optimize clearing when writing

Nick proposed years ago that tmpfs should avoid clearing its pages where
write will overwrite them with new data, as ramfs has long done.  But I
messed it up and just got bad data.  Tried again recently, it works
fine.

Here's time output for writing 4GiB 16 times on this Core i5 laptop:

before: real 0m21.169s user 0m0.028s sys 0m21.057s
        real 0m21.382s user 0m0.016s sys 0m21.289s
        real 0m21.311s user 0m0.020s sys 0m21.217s

after:  real 0m18.273s user 0m0.032s sys 0m18.165s
        real 0m18.354s user 0m0.020s sys 0m18.265s
        real 0m18.440s user 0m0.032s sys 0m18.337s

ramfs:  real 0m16.860s user 0m0.028s sys 0m16.765s
        real 0m17.382s user 0m0.040s sys 0m17.273s
        real 0m17.133s user 0m0.044s sys 0m17.021s

Yes, I have done perf reports, but they need more explanation than they
deserve: in summary, clear_page vanishes, its cache loading shifts into
copy_user_generic_unrolled; shmem_getpage_gfp goes down, and
surprisingly mark_page_accessed goes way up - I think because they are
respectively where the cache gets to be reloaded after being purged by
clear or copy.

Suggested-by: Nick Piggin <npiggin@gmail.com>
Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/shmem.c